The ancient Greek thinker and polymath Aristotle once deduced that the human heart comprises three chambers and regarded it as the most crucial organ within the body, regulating movement, sensation, and cognition.
In contemporary times, we understand that the human heart is comprised of four chambers and that the brain predominantly governs movement, sensation, and cognition. Nonetheless, Aristotle was astute in recognizing that the heart is an essential organ, circulating blood to other critical organs needed for their operations. When a serious condition such as heart failure occurs, the heart progressively loses its capacity to deliver sufficient blood and nutrients to other organs, impeding their functionality.
Researchers from MIT and Harvard Medical School have recently released an open-access study in Nature Communications Medicine, unveiling a non-invasive deep learning technique that evaluates electrocardiogram (ECG) signals to precisely anticipate a patient’s risk of experiencing heart failure. In a clinical trial, the model demonstrated accuracy comparable to the gold standard but less invasive methods, instilling hope for individuals at risk of heart failure. This condition has recently encountered a significant spike in mortality rates, especially among younger adults, possibly due to the rising occurrence of obesity and diabetes.
“This paper is the culmination of discussions I’ve had in various forums over the past several years,” states the paper’s lead author Collin Stultz, director of Harvard-MIT Program in Health Sciences and Technology and affiliate of the MIT Abdul Latif Jameel Clinic for Machine Learning in Health (Jameel Clinic). “The objective of this research is to identify individuals who are beginning to fall ill even before they display symptoms, allowing for early intervention to prevent hospitalization.”
Among the heart’s four chambers, two are atria and two are ventricles — the right side houses one atrium and one ventricle, while the left side mirrors this. In a healthy heart, these chambers function in a rhythmic harmony: oxygen-poor blood enters the heart via the right atrium. Upon contraction, the right atrium generates pressure that propels the blood into the right ventricle, from which it is pumped into the lungs for oxygenation. The oxygen-rich blood from the lungs then flows into the left atrium, which contracts to push the blood into the left ventricle. Subsequently, another contraction occurs, and the blood is expelled from the left ventricle through the aorta, circulating into veins that distribute it throughout the body.
“When left atrial pressures rise, the drainage of blood from the lungs into the left atrium is obstructed because it operates under higher pressure,” Stultz clarifies. In addition to holding a position as a professor of electrical engineering and computer science, Stultz practices as a cardiologist at Mass General Hospital (MGH). “The greater the pressure in the left atrium, the more pulmonary symptoms arise, such as shortness of breath. Given that the right side of the heart pumps blood through the pulmonary network to the lungs, heightened pressures in the left atrium result in increased pressures in the pulmonary circulation.”
The prevailing gold standard for assessing left atrial pressure is right heart catheterization (RHC), an invasive procedure involving a thin tube (catheter) attached to a pressure transmitter inserted into the right heart and pulmonary arteries. Medical professionals often opt to evaluate risk non-invasively prior to choosing RHC, by analyzing the patient’s weight, blood pressure, and heart rate.
However, Stultz believes these evaluations are somewhat crude, as highlighted by the fact that one in four heart failure patients is readmitted to the hospital within a month. “What we are looking for is a solution that provides information akin to that obtained from an invasive device, beyond a mere weight scale,” explains Stultz.
To obtain more in-depth information regarding a patient’s heart condition, doctors generally employ a 12-lead ECG, which requires 10 adhesive patches to be affixed to the patient and connected to a machine that gathers data from 12 distinct angles of the heart. Nevertheless, 12-lead ECG machines are only available in clinical facilities and are not primarily intended to assess heart failure risk.
In place of that, Stultz and his fellow researchers suggest a Cardiac Hemodynamic AI monitoring System (CHAIS), a deep neural network designed to analyze ECG data from a single lead — meaning the patient simply needs to wear one adhesive, commercially-available patch on their chest that can be utilized outside the hospital, disconnected from a machine.
To evaluate CHAIS against the existing gold standard, RHC, the researchers selected patients who were already scheduled for catheterization, inviting them to don the patch 24 to 48 hours before the procedure, although patients were instructed to remove the patch prior to catheterization. “When you get within an hour and a half [before the procedure], it’s 0.875, which is very, very accurate,” Stultz elaborates. “Thus, a measure from the device is equivalent and provides the same information as if you were catheterized in the next hour and a half.”
“Every cardiologist acknowledges the significance of left atrial pressure measurements in characterizing cardiac function and tailoring treatment strategies for heart failure patients,” remarks Aaron Aguirre SM ’03, PhD ’08, a cardiologist and critical care physician at MGH. “This study is critical as it provides a noninvasive method for estimating this vital clinical metric utilizing a widely accessible cardiac monitor.”
Aguirre, who obtained a PhD in medical engineering and medical physics at MIT, anticipates that with further clinical validation, CHAIS will prove beneficial in two primary domains: initially, it will assist in identifying patients who would gain the most from more invasive cardiac testing via RHC; and secondly, the technology could facilitate continuous monitoring and tracking of left atrial pressure in patients with heart disease. “A noninvasive and quantitative approach can enhance treatment strategies for patients both at home and in the hospital,” Aguirre states. “I am eager to observe the advancements the MIT team achieves next.”
Yet the advantages extend beyond patients alone — for individuals grappling with difficult-to-manage heart failure, it poses a challenge to prevent hospital readmissions without a permanent implant, consuming more space and time from an already strained and underfunded healthcare workforce.
The researchers are currently conducting another clinical trial utilizing CHAIS with MGH and Boston Medical Center that they aim to conclude soon for data analysis.
“In my opinion, the true potential of AI in healthcare is to deliver equitable, cutting-edge care to everyone, regardless of economic status, background, or location,” Stultz asserts. “This research is a significant move towards achieving this objective.”